Use check_same_thread=False by default for history sqlite db (#13886)

I had run into the following exception while trying to use IPython in a
thread:
```
Traceback (most recent call last):
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/interactiveshell.py", line 3745, in atexit_operations
    self._atexit_once()
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/interactiveshell.py", line 3728, in _atexit_once
    self.history_manager.end_session()
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/history.py", line 576, in end_session
    self.writeout_cache()
  File "/usr/local/lib/python3.11/dist-packages/decorator.py", line 232, in fun
    return caller(func, *(extras + args), **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/history.py", line 60, in only_when_enabled
    return f(self, *a, **kw)
           ^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/history.py", line 831, in writeout_cache
    self._writeout_input_cache(conn)
  File "/usr/local/lib/python3.11/dist-packages/IPython/core/history.py", line 812, in _writeout_input_cache
    with conn:
sqlite3.ProgrammingError: SQLite objects created in a thread can only be used in that same thread. The object was created in thread
 id 139673788811008 and this is thread id 139673823184704.
```
And discovered that an issue (#680) has been open for it since 2011.
Back in
[2012](https://github.com/ipython/ipython/issues/680#issuecomment-3444922),
it seems like the only reason not to fix it was that the parameter
`check_same_thread` was not documented, but now [it
is](https://docs.python.org/3.11/library/sqlite3.html#sqlite3.connect),
and it has been at least since
[3.6](https://docs.python.org/3.6/library/sqlite3.html#sqlite3.connect).

Note that according to the docs:
> check_same_thread
([bool](https://docs.python.org/3/library/functions.html#bool)) – If
True (default), only the creating thread may use the connection. If
False, the connection may be shared across multiple threads; if so,
write operations should be serialized by the user to avoid data
corruption.

But I don't think this is an issue here. The operations should be
synchronized by the user only *on the same connection object*, and if
I'm not mistaken, if two instances of IPython were started in separate
threads, they'd each have their own history manager with its own
connection. The exception above (and the one in the related issue) is
raised from `atexit`, when the main thread is running finalizations
registered by other threads. At that point, the operations are serial
anyway because they are executed in a single thread.

IPython/core/history.py

@@ -1,968 +1,972 b''
 """ History related magics and functionality """
 
 # Copyright (c) IPython Development Team.
 # Distributed under the terms of the Modified BSD License.
 
 
 import atexit
 import datetime
 from pathlib import Path
 import re
 import sqlite3
 import threading
 
 from traitlets.config.configurable import LoggingConfigurable
 from decorator import decorator
 from IPython.utils.decorators import undoc
 from IPython.paths import locate_profile
 from traitlets import (
     Any,
     Bool,
     Dict,
     Instance,
     Integer,
     List,
     Unicode,
     Union,
     TraitError,
     default,
     observe,
 )
 
 #-----------------------------------------------------------------------------
 # Classes and functions
 #-----------------------------------------------------------------------------
 
 @undoc
 class DummyDB(object):
     """Dummy DB that will act as a black hole for history.
 
     Only used in the absence of sqlite"""
     def execute(*args, **kwargs):
         return []
 
     def commit(self, *args, **kwargs):
         pass
 
     def __enter__(self, *args, **kwargs):
         pass
 
     def __exit__(self, *args, **kwargs):
         pass
 
 
 @decorator
 def only_when_enabled(f, self, *a, **kw):
     """Decorator: return an empty list in the absence of sqlite."""
     if not self.enabled:
         return []
     else:
         return f(self, *a, **kw)
 
 
 # use 16kB as threshold for whether a corrupt history db should be saved
 # that should be at least 100 entries or so
 _SAVE_DB_SIZE = 16384
 
 @decorator
 def catch_corrupt_db(f, self, *a, **kw):
     """A decorator which wraps HistoryAccessor method calls to catch errors from
     a corrupt SQLite database, move the old database out of the way, and create
     a new one.
 
     We avoid clobbering larger databases because this may be triggered due to filesystem issues,
     not just a corrupt file.
     """
     try:
         return f(self, *a, **kw)
     except (sqlite3.DatabaseError, sqlite3.OperationalError) as e:
         self._corrupt_db_counter += 1
         self.log.error("Failed to open SQLite history %s (%s).", self.hist_file, e)
         if self.hist_file != ':memory:':
             if self._corrupt_db_counter > self._corrupt_db_limit:
                 self.hist_file = ':memory:'
                 self.log.error("Failed to load history too many times, history will not be saved.")
             elif self.hist_file.is_file():
                 # move the file out of the way
                 base = str(self.hist_file.parent / self.hist_file.stem)
                 ext = self.hist_file.suffix
                 size = self.hist_file.stat().st_size
                 if size >= _SAVE_DB_SIZE:
                     # if there's significant content, avoid clobbering
                     now = datetime.datetime.now().isoformat().replace(':', '.')
                     newpath = base + '-corrupt-' + now + ext
                     # don't clobber previous corrupt backups
                     for i in range(100):
                         if not Path(newpath).exists():
                             break
                         else:
                             newpath = base + '-corrupt-' + now + (u'-%i' % i) + ext
                 else:
                     # not much content, possibly empty; don't worry about clobbering
                     # maybe we should just delete it?
                     newpath = base + '-corrupt' + ext
                 self.hist_file.rename(newpath)
                 self.log.error("History file was moved to %s and a new file created.", newpath)
             self.init_db()
             return []
         else:
             # Failed with :memory:, something serious is wrong
             raise
 
 
 class HistoryAccessorBase(LoggingConfigurable):
     """An abstract class for History Accessors """
 
     def get_tail(self, n=10, raw=True, output=False, include_latest=False):
         raise NotImplementedError
 
     def search(self, pattern="*", raw=True, search_raw=True,
                output=False, n=None, unique=False):
         raise NotImplementedError
 
     def get_range(self, session, start=1, stop=None, raw=True,output=False):
         raise NotImplementedError
 
     def get_range_by_str(self, rangestr, raw=True, output=False):
         raise NotImplementedError
 
 
 class HistoryAccessor(HistoryAccessorBase):
     """Access the history database without adding to it.
 
     This is intended for use by standalone history tools. IPython shells use
     HistoryManager, below, which is a subclass of this."""
 
     # counter for init_db retries, so we don't keep trying over and over
     _corrupt_db_counter = 0
      # after two failures, fallback on :memory:
     _corrupt_db_limit = 2
 
     # String holding the path to the history file
     hist_file = Union(
         [Instance(Path), Unicode()],
         help="""Path to file to use for SQLite history database.
 
         By default, IPython will put the history database in the IPython
         profile directory.  If you would rather share one history among
         profiles, you can set this value in each, so that they are consistent.
 
         Due to an issue with fcntl, SQLite is known to misbehave on some NFS
         mounts.  If you see IPython hanging, try setting this to something on a
         local disk, e.g::
 
             ipython --HistoryManager.hist_file=/tmp/ipython_hist.sqlite
 
         you can also use the specific value `:memory:` (including the colon
         at both end but not the back ticks), to avoid creating an history file.
 
         """,
     ).tag(config=True)
 
     enabled = Bool(True,
         help="""enable the SQLite history
 
         set enabled=False to disable the SQLite history,
         in which case there will be no stored history, no SQLite connection,
         and no background saving thread.  This may be necessary in some
         threaded environments where IPython is embedded.
         """,
     ).tag(config=True)
 
     connection_options = Dict(
         help="""Options for configuring the SQLite connection
 
         These options are passed as keyword args to sqlite3.connect
         when establishing database connections.
         """
     ).tag(config=True)
 
+    @default("connection_options")
+    def _default_connection_options(self):
+        return dict(check_same_thread=False)
+
     # The SQLite database
     db = Any()
     @observe('db')
     def _db_changed(self, change):
         """validate the db, since it can be an Instance of two different types"""
         new = change['new']
         connection_types = (DummyDB, sqlite3.Connection)
         if not isinstance(new, connection_types):
             msg = "%s.db must be sqlite3 Connection or DummyDB, not %r" % \
                     (self.__class__.__name__, new)
             raise TraitError(msg)
 
     def __init__(self, profile="default", hist_file="", **traits):
         """Create a new history accessor.
 
         Parameters
         ----------
         profile : str
             The name of the profile from which to open history.
         hist_file : str
             Path to an SQLite history database stored by IPython. If specified,
             hist_file overrides profile.
         config : :class:`~traitlets.config.loader.Config`
             Config object. hist_file can also be set through this.
         """
         super(HistoryAccessor, self).__init__(**traits)
         # defer setting hist_file from kwarg until after init,
         # otherwise the default kwarg value would clobber any value
         # set by config
         if hist_file:
             self.hist_file = hist_file
 
         try:
             self.hist_file
         except TraitError:
             # No one has set the hist_file, yet.
             self.hist_file = self._get_hist_file_name(profile)
 
         self.init_db()
 
     def _get_hist_file_name(self, profile='default'):
         """Find the history file for the given profile name.
 
         This is overridden by the HistoryManager subclass, to use the shell's
         active profile.
 
         Parameters
         ----------
         profile : str
             The name of a profile which has a history file.
         """
         return Path(locate_profile(profile)) / "history.sqlite"
 
     @catch_corrupt_db
     def init_db(self):
         """Connect to the database, and create tables if necessary."""
         if not self.enabled:
             self.db = DummyDB()
             return
 
         # use detect_types so that timestamps return datetime objects
         kwargs = dict(detect_types=sqlite3.PARSE_DECLTYPES|sqlite3.PARSE_COLNAMES)
         kwargs.update(self.connection_options)
         self.db = sqlite3.connect(str(self.hist_file), **kwargs)
         with self.db:
             self.db.execute(
                 """CREATE TABLE IF NOT EXISTS sessions (session integer
                             primary key autoincrement, start timestamp,
                             end timestamp, num_cmds integer, remark text)"""
             )
             self.db.execute(
                 """CREATE TABLE IF NOT EXISTS history
                     (session integer, line integer, source text, source_raw text,
                     PRIMARY KEY (session, line))"""
             )
             # Output history is optional, but ensure the table's there so it can be
             # enabled later.
             self.db.execute(
                 """CREATE TABLE IF NOT EXISTS output_history
                             (session integer, line integer, output text,
                             PRIMARY KEY (session, line))"""
             )
         # success! reset corrupt db count
         self._corrupt_db_counter = 0
 
     def writeout_cache(self):
         """Overridden by HistoryManager to dump the cache before certain
         database lookups."""
         pass
 
     ## -------------------------------
     ## Methods for retrieving history:
     ## -------------------------------
     def _run_sql(self, sql, params, raw=True, output=False, latest=False):
         """Prepares and runs an SQL query for the history database.
 
         Parameters
         ----------
         sql : str
             Any filtering expressions to go after SELECT ... FROM ...
         params : tuple
             Parameters passed to the SQL query (to replace "?")
         raw, output : bool
             See :meth:`get_range`
         latest : bool
             Select rows with max (session, line)
 
         Returns
         -------
         Tuples as :meth:`get_range`
         """
         toget = 'source_raw' if raw else 'source'
         sqlfrom = "history"
         if output:
             sqlfrom = "history LEFT JOIN output_history USING (session, line)"
             toget = "history.%s, output_history.output" % toget
         if latest:
             toget += ", MAX(session * 128 * 1024 + line)"
         this_querry = "SELECT session, line, %s FROM %s " % (toget, sqlfrom) + sql
         cur = self.db.execute(this_querry, params)
         if latest:
             cur = (row[:-1] for row in cur)
         if output:    # Regroup into 3-tuples, and parse JSON
             return ((ses, lin, (inp, out)) for ses, lin, inp, out in cur)
         return cur
 
     @only_when_enabled
     @catch_corrupt_db
     def get_session_info(self, session):
         """Get info about a session.
 
         Parameters
         ----------
         session : int
             Session number to retrieve.
 
         Returns
         -------
         session_id : int
             Session ID number
         start : datetime
             Timestamp for the start of the session.
         end : datetime
             Timestamp for the end of the session, or None if IPython crashed.
         num_cmds : int
             Number of commands run, or None if IPython crashed.
         remark : unicode
             A manually set description.
         """
         query = "SELECT * from sessions where session == ?"
         return self.db.execute(query, (session,)).fetchone()
 
     @catch_corrupt_db
     def get_last_session_id(self):
         """Get the last session ID currently in the database.
 
         Within IPython, this should be the same as the value stored in
         :attr:`HistoryManager.session_number`.
         """
         for record in self.get_tail(n=1, include_latest=True):
             return record[0]
 
     @catch_corrupt_db
     def get_tail(self, n=10, raw=True, output=False, include_latest=False):
         """Get the last n lines from the history database.
 
         Parameters
         ----------
         n : int
             The number of lines to get
         raw, output : bool
             See :meth:`get_range`
         include_latest : bool
             If False (default), n+1 lines are fetched, and the latest one
             is discarded. This is intended to be used where the function
             is called by a user command, which it should not return.
 
         Returns
         -------
         Tuples as :meth:`get_range`
         """
         self.writeout_cache()
         if not include_latest:
             n += 1
         cur = self._run_sql(
             "ORDER BY session DESC, line DESC LIMIT ?", (n,), raw=raw, output=output
         )
         if not include_latest:
             return reversed(list(cur)[1:])
         return reversed(list(cur))
 
     @catch_corrupt_db
     def search(self, pattern="*", raw=True, search_raw=True,
                output=False, n=None, unique=False):
         """Search the database using unix glob-style matching (wildcards
         * and ?).
 
         Parameters
         ----------
         pattern : str
             The wildcarded pattern to match when searching
         search_raw : bool
             If True, search the raw input, otherwise, the parsed input
         raw, output : bool
             See :meth:`get_range`
         n : None or int
             If an integer is given, it defines the limit of
             returned entries.
         unique : bool
             When it is true, return only unique entries.
 
         Returns
         -------
         Tuples as :meth:`get_range`
         """
         tosearch = "source_raw" if search_raw else "source"
         if output:
             tosearch = "history." + tosearch
         self.writeout_cache()
         sqlform = "WHERE %s GLOB ?" % tosearch
         params = (pattern,)
         if unique:
             sqlform += ' GROUP BY {0}'.format(tosearch)
         if n is not None:
             sqlform += " ORDER BY session DESC, line DESC LIMIT ?"
             params += (n,)
         elif unique:
             sqlform += " ORDER BY session, line"
         cur = self._run_sql(sqlform, params, raw=raw, output=output, latest=unique)
         if n is not None:
             return reversed(list(cur))
         return cur
 
     @catch_corrupt_db
     def get_range(self, session, start=1, stop=None, raw=True,output=False):
         """Retrieve input by session.
 
         Parameters
         ----------
         session : int
             Session number to retrieve.
         start : int
             First line to retrieve.
         stop : int
             End of line range (excluded from output itself). If None, retrieve
             to the end of the session.
         raw : bool
             If True, return untranslated input
         output : bool
             If True, attempt to include output. This will be 'real' Python
             objects for the current session, or text reprs from previous
             sessions if db_log_output was enabled at the time. Where no output
             is found, None is used.
 
         Returns
         -------
         entries
             An iterator over the desired lines. Each line is a 3-tuple, either
             (session, line, input) if output is False, or
             (session, line, (input, output)) if output is True.
         """
         if stop:
             lineclause = "line >= ? AND line < ?"
             params = (session, start, stop)
         else:
             lineclause = "line>=?"
             params = (session, start)
 
         return self._run_sql("WHERE session==? AND %s" % lineclause,
                                     params, raw=raw, output=output)
 
     def get_range_by_str(self, rangestr, raw=True, output=False):
         """Get lines of history from a string of ranges, as used by magic
         commands %hist, %save, %macro, etc.
 
         Parameters
         ----------
         rangestr : str
             A string specifying ranges, e.g. "5 ~2/1-4". If empty string is used,
             this will return everything from current session's history.
 
             See the documentation of :func:`%history` for the full details.
 
         raw, output : bool
             As :meth:`get_range`
 
         Returns
         -------
         Tuples as :meth:`get_range`
         """
         for sess, s, e in extract_hist_ranges(rangestr):
             for line in self.get_range(sess, s, e, raw=raw, output=output):
                 yield line
 
 
 class HistoryManager(HistoryAccessor):
     """A class to organize all history-related functionality in one place.
     """
     # Public interface
 
     # An instance of the IPython shell we are attached to
     shell = Instance('IPython.core.interactiveshell.InteractiveShellABC',
                      allow_none=True)
     # Lists to hold processed and raw history. These start with a blank entry
     # so that we can index them starting from 1
     input_hist_parsed = List([""])
     input_hist_raw = List([""])
     # A list of directories visited during session
     dir_hist = List()
     @default('dir_hist')
     def _dir_hist_default(self):
         try:
             return [Path.cwd()]
         except OSError:
             return []
 
     # A dict of output history, keyed with ints from the shell's
     # execution count.
     output_hist = Dict()
     # The text/plain repr of outputs.
     output_hist_reprs = Dict()
 
     # The number of the current session in the history database
     session_number = Integer()
 
     db_log_output = Bool(False,
         help="Should the history database include output? (default: no)"
     ).tag(config=True)
     db_cache_size = Integer(0,
         help="Write to database every x commands (higher values save disk access & power).\n"
         "Values of 1 or less effectively disable caching."
     ).tag(config=True)
     # The input and output caches
     db_input_cache = List()
     db_output_cache = List()
 
     # History saving in separate thread
     save_thread = Instance('IPython.core.history.HistorySavingThread',
                            allow_none=True)
     save_flag = Instance(threading.Event, allow_none=True)
 
     # Private interface
     # Variables used to store the three last inputs from the user.  On each new
     # history update, we populate the user's namespace with these, shifted as
     # necessary.
     _i00 = Unicode(u'')
     _i = Unicode(u'')
     _ii = Unicode(u'')
     _iii = Unicode(u'')
 
     # A regex matching all forms of the exit command, so that we don't store
     # them in the history (it's annoying to rewind the first entry and land on
     # an exit call).
     _exit_re = re.compile(r"(exit|quit)(\s*\(.*\))?$")
 
     def __init__(self, shell=None, config=None, **traits):
         """Create a new history manager associated with a shell instance.
         """
         super(HistoryManager, self).__init__(shell=shell, config=config,
             **traits)
         self.save_flag = threading.Event()
         self.db_input_cache_lock = threading.Lock()
         self.db_output_cache_lock = threading.Lock()
 
         try:
             self.new_session()
         except sqlite3.OperationalError:
             self.log.error("Failed to create history session in %s. History will not be saved.",
                 self.hist_file, exc_info=True)
             self.hist_file = ':memory:'
 
         if self.enabled and self.hist_file != ':memory:':
             self.save_thread = HistorySavingThread(self)
             self.save_thread.start()
 
     def _get_hist_file_name(self, profile=None):
         """Get default history file name based on the Shell's profile.
 
         The profile parameter is ignored, but must exist for compatibility with
         the parent class."""
         profile_dir = self.shell.profile_dir.location
         return Path(profile_dir) / "history.sqlite"
 
     @only_when_enabled
     def new_session(self, conn=None):
         """Get a new session number."""
         if conn is None:
             conn = self.db
 
         with conn:
             cur = conn.execute(
                 """INSERT INTO sessions VALUES (NULL, ?, NULL,
                             NULL, '') """,
                 (datetime.datetime.now(),),
             )
             self.session_number = cur.lastrowid
 
     def end_session(self):
         """Close the database session, filling in the end time and line count."""
         self.writeout_cache()
         with self.db:
             self.db.execute("""UPDATE sessions SET end=?, num_cmds=? WHERE
                             session==?""", (datetime.datetime.now(),
                             len(self.input_hist_parsed)-1, self.session_number))
         self.session_number = 0
 
     def name_session(self, name):
         """Give the current session a name in the history database."""
         with self.db:
             self.db.execute("UPDATE sessions SET remark=? WHERE session==?",
                             (name, self.session_number))
 
     def reset(self, new_session=True):
         """Clear the session history, releasing all object references, and
         optionally open a new session."""
         self.output_hist.clear()
         # The directory history can't be completely empty
         self.dir_hist[:] = [Path.cwd()]
 
         if new_session:
             if self.session_number:
                 self.end_session()
             self.input_hist_parsed[:] = [""]
             self.input_hist_raw[:] = [""]
             self.new_session()
 
     # ------------------------------
     # Methods for retrieving history
     # ------------------------------
     def get_session_info(self, session=0):
         """Get info about a session.
 
         Parameters
         ----------
         session : int
             Session number to retrieve. The current session is 0, and negative
             numbers count back from current session, so -1 is the previous session.
 
         Returns
         -------
         session_id : int
             Session ID number
         start : datetime
             Timestamp for the start of the session.
         end : datetime
             Timestamp for the end of the session, or None if IPython crashed.
         num_cmds : int
             Number of commands run, or None if IPython crashed.
         remark : unicode
             A manually set description.
         """
         if session <= 0:
             session += self.session_number
 
         return super(HistoryManager, self).get_session_info(session=session)
 
     @catch_corrupt_db
     def get_tail(self, n=10, raw=True, output=False, include_latest=False):
         """Get the last n lines from the history database.
 
         Most recent entry last.
 
         Completion will be reordered so that that the last ones are when
         possible from current session.
 
         Parameters
         ----------
         n : int
             The number of lines to get
         raw, output : bool
             See :meth:`get_range`
         include_latest : bool
             If False (default), n+1 lines are fetched, and the latest one
             is discarded. This is intended to be used where the function
             is called by a user command, which it should not return.
 
         Returns
         -------
         Tuples as :meth:`get_range`
         """
         self.writeout_cache()
         if not include_latest:
             n += 1
         # cursor/line/entry
         this_cur = list(
             self._run_sql(
                 "WHERE session == ? ORDER BY line DESC LIMIT ?  ",
                 (self.session_number, n),
                 raw=raw,
                 output=output,
             )
         )
         other_cur = list(
             self._run_sql(
                 "WHERE session != ? ORDER BY session DESC, line DESC LIMIT ?",
                 (self.session_number, n),
                 raw=raw,
                 output=output,
             )
         )
 
         everything = this_cur + other_cur
 
         everything = everything[:n]
 
         if not include_latest:
             return list(everything)[:0:-1]
         return list(everything)[::-1]
 
     def _get_range_session(self, start=1, stop=None, raw=True, output=False):
         """Get input and output history from the current session. Called by
         get_range, and takes similar parameters."""
         input_hist = self.input_hist_raw if raw else self.input_hist_parsed
 
         n = len(input_hist)
         if start < 0:
             start += n
         if not stop or (stop > n):
             stop = n
         elif stop < 0:
             stop += n
 
         for i in range(start, stop):
             if output:
                 line = (input_hist[i], self.output_hist_reprs.get(i))
             else:
                 line = input_hist[i]
             yield (0, i, line)
 
     def get_range(self, session=0, start=1, stop=None, raw=True,output=False):
         """Retrieve input by session.
 
         Parameters
         ----------
         session : int
             Session number to retrieve. The current session is 0, and negative
             numbers count back from current session, so -1 is previous session.
         start : int
             First line to retrieve.
         stop : int
             End of line range (excluded from output itself). If None, retrieve
             to the end of the session.
         raw : bool
             If True, return untranslated input
         output : bool
             If True, attempt to include output. This will be 'real' Python
             objects for the current session, or text reprs from previous
             sessions if db_log_output was enabled at the time. Where no output
             is found, None is used.
 
         Returns
         -------
         entries
             An iterator over the desired lines. Each line is a 3-tuple, either
             (session, line, input) if output is False, or
             (session, line, (input, output)) if output is True.
         """
         if session <= 0:
             session += self.session_number
         if session==self.session_number:          # Current session
             return self._get_range_session(start, stop, raw, output)
         return super(HistoryManager, self).get_range(session, start, stop, raw,
                                                      output)
 
     ## ----------------------------
     ## Methods for storing history:
     ## ----------------------------
     def store_inputs(self, line_num, source, source_raw=None):
         """Store source and raw input in history and create input cache
         variables ``_i*``.
 
         Parameters
         ----------
         line_num : int
             The prompt number of this input.
         source : str
             Python input.
         source_raw : str, optional
             If given, this is the raw input without any IPython transformations
             applied to it.  If not given, ``source`` is used.
         """
         if source_raw is None:
             source_raw = source
         source = source.rstrip('\n')
         source_raw = source_raw.rstrip('\n')
 
         # do not store exit/quit commands
         if self._exit_re.match(source_raw.strip()):
             return
 
         self.input_hist_parsed.append(source)
         self.input_hist_raw.append(source_raw)
 
         with self.db_input_cache_lock:
             self.db_input_cache.append((line_num, source, source_raw))
             # Trigger to flush cache and write to DB.
             if len(self.db_input_cache) >= self.db_cache_size:
                 self.save_flag.set()
 
         # update the auto _i variables
         self._iii = self._ii
         self._ii = self._i
         self._i = self._i00
         self._i00 = source_raw
 
         # hackish access to user namespace to create _i1,_i2... dynamically
         new_i = '_i%s' % line_num
         to_main = {'_i': self._i,
                    '_ii': self._ii,
                    '_iii': self._iii,
                    new_i : self._i00 }
 
         if self.shell is not None:
             self.shell.push(to_main, interactive=False)
 
     def store_output(self, line_num):
         """If database output logging is enabled, this saves all the
         outputs from the indicated prompt number to the database. It's
         called by run_cell after code has been executed.
 
         Parameters
         ----------
         line_num : int
             The line number from which to save outputs
         """
         if (not self.db_log_output) or (line_num not in self.output_hist_reprs):
             return
         output = self.output_hist_reprs[line_num]
 
         with self.db_output_cache_lock:
             self.db_output_cache.append((line_num, output))
         if self.db_cache_size <= 1:
             self.save_flag.set()
 
     def _writeout_input_cache(self, conn):
         with conn:
             for line in self.db_input_cache:
                 conn.execute("INSERT INTO history VALUES (?, ?, ?, ?)",
                                 (self.session_number,)+line)
 
     def _writeout_output_cache(self, conn):
         with conn:
             for line in self.db_output_cache:
                 conn.execute("INSERT INTO output_history VALUES (?, ?, ?)",
                                 (self.session_number,)+line)
 
     @only_when_enabled
     def writeout_cache(self, conn=None):
         """Write any entries in the cache to the database."""
         if conn is None:
             conn = self.db
 
         with self.db_input_cache_lock:
             try:
                 self._writeout_input_cache(conn)
             except sqlite3.IntegrityError:
                 self.new_session(conn)
                 print("ERROR! Session/line number was not unique in",
                       "database. History logging moved to new session",
                                                 self.session_number)
                 try:
                     # Try writing to the new session. If this fails, don't
                     # recurse
                     self._writeout_input_cache(conn)
                 except sqlite3.IntegrityError:
                     pass
             finally:
                 self.db_input_cache = []
 
         with self.db_output_cache_lock:
             try:
                 self._writeout_output_cache(conn)
             except sqlite3.IntegrityError:
                 print("!! Session/line number for output was not unique",
                       "in database. Output will not be stored.")
             finally:
                 self.db_output_cache = []
 
 
 class HistorySavingThread(threading.Thread):
     """This thread takes care of writing history to the database, so that
     the UI isn't held up while that happens.
 
     It waits for the HistoryManager's save_flag to be set, then writes out
     the history cache. The main thread is responsible for setting the flag when
     the cache size reaches a defined threshold."""
     daemon = True
     stop_now = False
     enabled = True
     def __init__(self, history_manager):
         super(HistorySavingThread, self).__init__(name="IPythonHistorySavingThread")
         self.history_manager = history_manager
         self.enabled = history_manager.enabled
         atexit.register(self.stop)
 
     @only_when_enabled
     def run(self):
         # We need a separate db connection per thread:
         try:
             self.db = sqlite3.connect(
                 str(self.history_manager.hist_file),
                 **self.history_manager.connection_options,
             )
             while True:
                 self.history_manager.save_flag.wait()
                 if self.stop_now:
                     self.db.close()
                     return
                 self.history_manager.save_flag.clear()
                 self.history_manager.writeout_cache(self.db)
         except Exception as e:
             print(("The history saving thread hit an unexpected error (%s)."
                    "History will not be written to the database.") % repr(e))
 
     def stop(self):
         """This can be called from the main thread to safely stop this thread.
 
         Note that it does not attempt to write out remaining history before
         exiting. That should be done by calling the HistoryManager's
         end_session method."""
         self.stop_now = True
         self.history_manager.save_flag.set()
         self.join()
 
 
 # To match, e.g. ~5/8-~2/3
 range_re = re.compile(r"""
 ((?P<startsess>~?\d+)/)?
 (?P<start>\d+)?
 ((?P<sep>[\-:])
  ((?P<endsess>~?\d+)/)?
  (?P<end>\d+))?
 $""", re.VERBOSE)
 
 
 def extract_hist_ranges(ranges_str):
     """Turn a string of history ranges into 3-tuples of (session, start, stop).
 
     Empty string results in a `[(0, 1, None)]`, i.e. "everything from current
     session".
 
     Examples
     --------
     >>> list(extract_hist_ranges("~8/5-~7/4 2"))
     [(-8, 5, None), (-7, 1, 5), (0, 2, 3)]
     """
     if ranges_str == "":
         yield (0, 1, None)  # Everything from current session
         return
 
     for range_str in ranges_str.split():
         rmatch = range_re.match(range_str)
         if not rmatch:
             continue
         start = rmatch.group("start")
         if start:
             start = int(start)
             end = rmatch.group("end")
             # If no end specified, get (a, a + 1)
             end = int(end) if end else start + 1
         else:  # start not specified
             if not rmatch.group('startsess'):  # no startsess
                 continue
             start = 1
             end = None  # provide the entire session hist
 
         if rmatch.group("sep") == "-":       # 1-3 == 1:4 --> [1, 2, 3]
             end += 1
         startsess = rmatch.group("startsess") or "0"
         endsess = rmatch.group("endsess") or startsess
         startsess = int(startsess.replace("~","-"))
         endsess = int(endsess.replace("~","-"))
         assert endsess >= startsess, "start session must be earlier than end session"
 
         if endsess == startsess:
             yield (startsess, start, end)
             continue
         # Multiple sessions in one range:
         yield (startsess, start, None)
         for sess in range(startsess+1, endsess):
             yield (sess, 1, None)
         yield (endsess, 1, end)
 
 
 def _format_lineno(session, line):
     """Helper function to format line numbers properly."""
     if session == 0:
         return str(line)
     return "%s#%s" % (session, line)